Sri Harsha Vadlamudi

Advisors
Kai Klede (M.Sc.),  Daniel Fährmann (M.Sc.) (Fraunhofer IGD), Prof. Dr. Björn Eskofier

Duration
11 / 2021 – 05 / 2022

Abstract
In urban transportation networks, traffic anomalies are common and have a detrimental impact on traffic efficiency, travel time, and air pollution. When traffic accidents, traffic congestion, and huge gatherings and activities, such as construction, occur on a road network, the traffic flow becomes abnormal. As a result, the detection of traffic anomalies is critical for traffic management and is important to focus in transportation research. GPS-equipped taxis can be considered as extensive sensors and the large-scale digital traces that are source to disclose many unknown details about the city dynamics and human behaviours.[3].

To achieve the objective, we start with an analysis of what kind of anomalies we can detect based on the trajectory data. This anomaly detection can be helpful in finding out situations that affect the traffic infrastructure in the city and is a vital task to ensure safety and security of the inhabitants and the infrastructure itself. These anomalous situations can be anything out of the ordinary which occur due to extreme weather conditions, accidents or crowd movements in the transportation system.

We build an anomaly detector based on graph neural networks to detect different types of anomalies which helps us to reduce the issues in the transportation structure. Transportation networks can be naturally represented in a graph-like structure where the nodes can be locations and the edges can be spatial relationships between them. That is the main reason to use graph neural networks. [1].

The evaluation proposed work will be performed on real world data collected from sensor sources in modern smart cities. This research work provides a support to the project funded by the German Federal Ministry of Education and Research and the Hessian Ministry of Higher Education, Research, Science and the Arts within their joint support of the National Research Center for Applied Cybersecurity ATHENE. A duration of tentatively six months is required to work on the thesis with a previous review of literature.

References:
[1] Ma, Xiaoxiao and Wu, Jia and Xue, Shan and Yang, Jian and Zhou, Chuan and Sheng, Quan Z and Xiong, Hui and Akoglu, Leman.:A Comprehensive Survey on Graph Anomaly Detection with Deep Learning. IEEE Transactions on Knowledge and Data Engineering.
[2] Kang, Zhao and Peng, Chong and Cheng, Qiang and Liu, Xinwang and Peng, Xi and Xu, Zenglin and Tian, Ling.: Structured graph learning for clustering and semi-supervised classification. Pattern Recognition, vol. 110, p. 107627, Feb. 2021.
[3] Deng, Ailin and Hooi, Bryan.: Graph neural network-based anomaly detection in multivariate time series. Proceedings of the AAAI Conference on Artificial Intelligence.